Interaction between neuronal encoding and population dynamics during categorization task switching in parietal...

Area of Science:

• Cognitive Neuroscience
• Primate Behavior
• Computational Neuroscience

Background:

• Categorization is a key cognitive process in primates.
• Neural encoding of categories across tasks with varying demands is not fully understood.

Purpose of the Study:

• Investigate how neural encoding and dynamics support categorization tasks with different cognitive and motor demands.
• Compare neural dynamics in primate parietal cortex with recurrent neural networks (RNNs) trained on similar tasks.

Main Methods:

• Recorded neural activity from primate parietal cortex during two distinct categorization tasks: one-interval categorization (OIC) and delayed match-to-category (DMC).
• Developed and trained RNNs on the same OIC and DMC tasks to model neural dynamics.
• Analyzed neuronal and RNN activity for category encoding properties and task-dependent dynamics.

Main Results:

• Neuronal category encoding generalized across both OIC and DMC tasks.
• Categorical encoding was more binary-like in the DMC task compared to a more graded representation in the OIC task.
• RNN analysis supported the hypothesis that attractor dynamics compress graded feature encoding into binary-like representations in the DMC task.

Conclusions:

• Neural representations of categories adapt to task demands, exhibiting distinct encoding properties (graded vs. binary-like).
• Recurrent neural networks with attractor dynamics can model the observed compression of information, suggesting a potential mechanism for task-specific neural encoding.